Crate mnemonist_quant 

TurboQuant: online vector quantization with near-optimal distortion rate.

Implements the algorithms from TurboQuant (arXiv:2504.19874):

• TurboQuantMse — MSE-optimal quantizer using random rotation + Lloyd-Max codebooks
• TurboQuantProd — unbiased inner-product quantizer (MSE + QJL residual)
• CompressedEmbeddingStore — binary storage format for quantized embeddings

Re-exports

pub use codebook::Codebook;

pub use mse::QuantizedVector;

pub use mse::TurboQuantMse;

pub use prod::QuantizedProdVector;

pub use prod::TurboQuantProd;

pub use qjl::QjlTransform;

pub use rotation::Rotation;

pub use store::CompressedEmbeddingStore;

Modules

codebook

Precomputed Lloyd-Max optimal scalar quantizer codebooks.

mse

TurboQuant_mse: MSE-optimal vector quantizer (Algorithm 1).

pack

Bit-packing utilities for storing b-bit indices compactly.

prod

TurboQuant_prod: unbiased inner-product quantizer (Algorithm 2).

qjl

Quantized Johnson-Lindenstrauss (QJL) 1-bit transform.

rotation

Random orthogonal rotation matrix for TurboQuant.

store

Compressed embedding store using TurboQuant quantization.

Enums

QuantError

Errors specific to quantization operations.